How to rank in ChatGPT

What ChatGPT actually does when it answers

When a question needs current information, ChatGPT runs a web search, retrieves a small set of pages, reads chunks of them, and writes an answer citing the ones it used. The retrieval step is closer to a search engine than most people expect — pages that are already indexed and already rank reasonably well are the ones that enter the candidate pool.

The selection step is different. Out of the retrieved candidates, the model favours pages that state a checkable claim in a self-contained sentence, because those are the safest to quote. A page that ranks third on Google but hedges every sentence will lose to a page that ranks eighth and says something concrete.

The five things that decide whether you get cited

In practice, five levers move the needle, in roughly this order of impact:

• Crawlability for AI agents. GPTBot and OAI-SearchBot must be allowed in robots.txt, and your pages must return real HTML without requiring JavaScript. A JavaScript-only single-page app is invisible to most of them.
• Entity clarity. The model needs to know what your brand is, in one sentence, in the same words, everywhere. Organization schema, a plain about page, and consistent naming across your site, LinkedIn and any directory listing.
• Answer-first structure. A question as the H2, then a 40-80 word direct answer directly beneath it. That block is the unit the model lifts.
• Checkable facts. Numbers, dates, named sources, specific product behaviour. 'Setup takes about 12 minutes' is citable. 'Setup is fast' is not.
• Topical depth. One thorough page on a question beats five thin ones. Retrieval rewards the page that covers the whole question, including the objections.

How to measure it without guessing

The mistake most people make is checking once, in their own ChatGPT window, and drawing a conclusion. Answers vary between runs, between accounts, and between the search-enabled and non-search modes. A single check tells you almost nothing.

What works is a fixed probe set: 20 to 40 questions a real buyer would ask in your category, run against the same models on a schedule, recording whether you were mentioned, whether you were cited with a link, and which competitor took the slot when you were not. Over four to six weeks that produces a trend you can act on.

ThoughtForge runs exactly that loop across ChatGPT, Claude, Perplexity and Gemini, and links each gap to the specific page that caused it.

The order to fix things in

Fix entity clarity before you publish anything new. If the model is unsure what your company is, more content just adds noise. Get the one-sentence description consistent, add Organization and Person schema, and make sure your about page states plainly what you sell and to whom.

Then take the three questions where you were omitted most often and write one genuinely thorough page per question, answer-first. Re-probe two weeks later. If the page is retrieved but not cited, the problem is quotability; if it is not retrieved at all, the problem is indexing or authority.

What we measured on real domains

Everything below comes from ThoughtForge runs against live domains, not from a study we read. Each run scores three things out of 100: identity (can an engine say who you are without contradicting itself), durability (does your brand survive a multi-turn conversation), and retrieval (can your pages be chunked into something quotable).

Across every domain we have scanned, one number is consistently the worst, and it is not the one people optimise for. Durability sits at or near zero almost everywhere. Brands that get named in turn one disappear by turn three, because nothing on the site gives the model a reason to keep holding onto them.

• thoughtforge.app, July: identity 16, durability 16, retrieval 49, overall 27.
• thoughtforge.app, September, after entity and schema work: identity 52, durability 0, retrieval 48, overall 33.
• A services domain (megap•••••••••••.com.au): identity 28, durability 0, retrieval 47, overall 25.
• An industry association (propertyasso•••••••.com.au): identity 4, durability 0, retrieval 50, overall 18.
• A consultancy (obse•••••••••.com): identity 40 in July, 52 in September; overall 30 to 35.

What actually moved the numbers

Identity is the score that responds fastest. On thoughtforge.app it went from 16 to 52 in about six weeks, and the work was unglamorous: one Organization schema block that agreed with itself, one description of the product used in the same words on the site, in the schema and on every third-party listing, and crawler access opened to GPTBot, ClaudeBot and PerplexityBot.

Retrieval barely moved — 49 to 48 — because retrieval is about page structure, not metadata, and we had not restructured the pages yet. That is the honest shape of this work: metadata fixes buy you identity, and only rewriting the pages buys you retrieval.

Durability moved backwards, from 16 to 0. Multi-turn persistence is the hardest of the three and the one almost nobody is working on. If your brand only needs to be named once, ignore it. If you are in a considered purchase where the buyer asks the assistant four or five follow-ups, it is the whole game.

What this means if you are starting today

Do not start with content volume. On every domain we scanned, the ceiling was set by identity and retrieval, and publishing more pages against a confused identity just spreads the confusion. Fix who you are first, then make three pages genuinely quotable, then measure again before writing anything new.

Expect the second measurement to be disappointing in one dimension. That is useful information, not failure — it tells you which lever you have not pulled yet.

A week-by-week sequence that works

Week one is measurement and access. Confirm GPTBot and OAI-SearchBot can fetch your pages, confirm those pages return real HTML rather than an empty shell, and take a baseline score so later changes can be attributed to something. Skipping the baseline is the most common mistake we see; without it every later change is a matter of opinion.

Week two is identity. Write one sentence that says what you are, who it is for and what makes it different, then use that exact sentence in your Organization schema, your about page, your LinkedIn description and every directory listing. Contradiction is what keeps identity scores low, not absence of information.

Week three is the three pages that matter most — usually your primary service page, your pricing page and the one guide that already gets impressions. Rewrite them answer-first. Week four, measure again, and expect identity to have moved more than retrieval.

What does not work, based on what we have tried

Publishing more pages against a weak identity. If the model cannot say what you are, twenty more articles give it twenty more chances to describe you inconsistently.

Stuffing brand names into copy. Answer engines quote passages that answer a question; a paragraph that mentions you four times reads like an advertisement and gets skipped.

Treating llms.txt or FAQ markup as the fix. Both are cheap and worth doing, and neither moved a single score for us on their own.

• Do: state checkable facts with numbers and dates.
• Do: keep one description of yourself everywhere.
• Do not: hide your primary answer behind JavaScript.
• Do not: publish ten thin pages instead of one thorough one.

Frequently asked questions

Can you pay to appear in ChatGPT answers?

No. As of 2026 there is no paid placement inside ChatGPT's cited sources. Appearing is a function of being retrievable, relevant and quotable.

How long does it take to start appearing?

If the page is already indexed by Google and Bing, changes to structure and quotability can show up in probes within two to four weeks. New pages take longer because they must be indexed first.

Does ChatGPT use Google's index?

ChatGPT's search uses its own crawling plus third-party search infrastructure. In practice, pages that Google and Bing have indexed and rank are far more likely to be retrieved.

Do I need to block or allow GPTBot?

Allow it if you want to be cited. Blocking GPTBot removes your pages from the candidate pool entirely.

Related guides

• AI search visibility — AI search visibility is how often answer engines mention and cite your brand. How to measure it properly, what moves it, and what to ignore.
• AEO vs SEO vs GEO — SEO wins rankings, AEO wins the answer, GEO wins the citation. What each term actually means, how they overlap, and how to measure all three.
• What is llms.txt? — llms.txt is a plain-markdown file that tells AI assistants what your site is and where to read more. What it does, what it doesn't, and how to write one.